Social Mixing and Business Activity in Philadelphia
A report on mobility during the coronavirus pandemic
Andrew Renninger and Ken Steif
1 Introduction
As the coronavirus spread in Philadelphia this year, travel patterns changed. With this shock to human mobility came another to business activity, as far fewer of us traveled to work, amenities or other pastimes. In order to understand the consequences of these changes, we use mobile phone GPS data to explore how Philadelphians changed their travel patterns before and during the pandemic. This analysis will explore general trends before examining night life—the restaurants and bars that both provide jobs and support vibrant streets—and certain bellwether industries, like those with office work, that doubtless provide a foundation for such night life.
With the goal of understanding the time-space patterns of resident movement in Philadelphia, the following section presents data from SafeGraph, a provider of such records. Note that SafeGraph collects data on a representative sample (10%) of the population across the country, so our indicators are not the true number of visits or journeys, but a slice. The data contain the terms defined in figure 1.1: the number of visitors is the count of devices flowing to a point of interest—be it from a given Census Block Group or total—while a connection is an origin-destination line between a Census Block Group and a point of interest, regardless of its weight. Our concern here is to track the flow of visitors from an origin location to a destination location in order to map flows over time and space.
1.1 Key definitions
Figure 1.2 below maps these origin-destination flows as network connections to show the extent to which areas throughout Philadelphia are connected to one another. Changes over time reveal how much the network thins out as the pandemic grows.
1.2 Mobility network
Each visit is a mobile device entering into a point of interest; these include parks and museums, restaurats and bars, or offices and hospitals. In figure 1.3 we map the distribution of these venues and businesses for context. We classify each point of interest by its description, which SafeGraph provides. 1 We can see that most businesses cluster in Center City or nearby but no businesses cluster more than restaurants and bars.
1.3 Business landscape
This analysis comprises different spatial scales, Citywide, Neighborhood and Point of Interest. We can look globally, across the city, to explore trends throughout; we can also think locally, dividing the city up into cells or neighborhoods to probe variations within the city. Finally, we can look at individual businesses or veneues. Below, we attempt to understand patterns at each scale.
2 Citywide analysis
In this section we explore trends and relationships manifest most strongly at the global level, across the city. Best described by this focus on the whole over its part are how certain brands and industries are performing, regardless of location, and how certain variables predict changes to activity and mobility in Philadelphia. We see how visitation is changing across time by tracking visits across brands. Figure 2.1 shows that brands associated with necessities (Target and Walmart) saw comparably less of a decline than others, along with fast food restaurants, which one might expect in a time of constrained budgets. The map shows the locations of brands for context.
2.1 Brand performance
| Rankings | Locations |
|---|---|
 |
 |
Figure 2.2, which ranks each brand by the number of visitors it received and animates this change through the pandemic. Dollar stores rise gradually throughout the year, an expected change as residents both need more home goods and need to save money; another important shift is away from non-essential retail towards essential businesses like pharmacies. Starbucks and Wawa occupy top spots for the first several weeks of the year but when the shelter-in-place order sets in, patronage immediately collapses and they are replaced in the ranks by essential shops RiteAid and ShopRite.
2.2 Changing fortunes
In figure 2.1 we aggregate by use, grouping by classes like leisure (restaurants and bars) and tourism (museums and theaters). The pandemic had distinct effects on each class, but particularly leisure and other; other includes offices which also explains the steep fall. Interestingly, tourism is recovering while shops and grocers are not, perhaps as many switch to digital commerce.
2.3 Industry trends
Changing mobility may influence or exacerbate existing problems in Philadelphia, notably around equity and integration. Philadelphia still shows patterns of concentrated poverty, segregated housing and isolated pockets of prosperty; the pandemic could produce deeper disparities. One risk is that communities of color and low income neighborhoods will not be able to socially distance in the same capacity as affluent communities. The data, however, do not give a clear signal. Below we plot the relationship between outflows—individuals visiting points of interest from a given tract—and key predictors: tract income and the percentage of the tract that is African American. (Tracts allow for better demographic estimates.) The story is clear for income, as during the critical month of April few poor communities could afford to shelter in place, but hazy for race.
2.4 Race and travel
The pandemic appears to have flattened an existing relationship between race and mobility: in early days of the pandemic, communities of color were more likely to receive visitors from the rest of the city, a pattern that held for peak months of spread, but this relationship weakens as more predominately white communities gained visitors in July and August. When we plot the same travel patterns against income, we see that wealthy communites are well below their baseline visits, perhaps because many of the restaurant clusters are in relatively affluent areas. While poor communites are more likely to have recovered to baseline but some of the poorest areas are still lagging behind, more in line with wealthy ones.
2.5 Income and travel
The story is similar when we look at inflows, which we document in the appendix. This suggests that poor and minority areas have remained comparable active during the pandemic, which may come with heightened exposure alongside economic stability.
3 Neighborhood variation
Perhaps more valuable than probing individual points of interest is aggregating by areal units, which we do in this section. These allow us to see how visits in particular are changing throughout the pandemic in different parts of Philadelphia. Philadelphia has roughly 150 neighborhoods (we use definitions from local firm Azavea) and each responded differently to the pandemic. We explore trends across neighborhoods in figure 3.1; neighborhoods dominated by office work, like the Navy Yard along with Logan Square and Center City, saw precipitous declines in foot traffic, but those with strong amenities and residential communities have recovered. This suggests that demand for food, drink, and shopping may be shifting away from the core. (Note: see the appendix for larger tables.)
3.1 Neighborhood trends
| Best | Worst |
|---|---|
 |
 |
Using a regular grid, in figure 3.2, we aggregate to a grid of 500 meter cells to see how this manifests across space. We are still aggregating from points of interest, so this is visits to businesses, parks, museums and the like, but by tile; this does not include visits to the particular patch of land without setting foot in a point of interest. The city hollowed out during the worst months of the pandemic but the Old City, Center City, University City axis still appears to have pockets of thriving activity in these maps.
3.2 Variation by time and geography
Businesses cluster together and we can explore the strength of this phenomenon by looking at commercial corridors, of which the city has designated roughly 280. Looking at night life in figure 3.3, the largest are Market West and Market East, on either side of city hall, with 1712 and 1263 restaurants and bars respectively, following by Old City at 654 and another in University City with 493: most of the business activity is concentrated in a few locales.
3.3 Commercial corridors
When we plot trends in these clusters over time, it is clear that many of the most successful areas are toward the periphery, perhaps dormitory communities supported by remote work, and several of the least successful are situated in the core. Notably among the worst performers are the two central corridors, which depend on office work, and the Sports Complex, which saw sports leagues take measures of protect players and ban fans early on—and many of these restrictions are still in place. Peripheral plazas like Oxford and Levick, home to a supermarket, and City and Haverford are among the best.
3.4 Corridor trends
| Best | Worst |
|---|---|
 |
 |
In figure 3.5 we plot distance to City Hall—geographic centrality—against visitation. The relationship between centrality —and activity is negligible, but the negative relationship between distance to City Hall and average visits suggests that activity clusters in the core while the positive relationship between distance and change in activity indicates that the core had farther to fall.
3.5 Explanatory variables
4 Points of Interest in focus
This section looks at individual points of interest, how they perform over time and whether or not we can identify certain bellwether businesses within the city. These cases can provide further insight into how the pandemic is changing mobility. We start by looking at the network of connections across the city. Drawing a line between each origin (neighborhood) and destination (point of interest), there is a dense web—a nearly saturated graph where all neighborhoods send visitors to all other corners of the city. This web becomes sparser as the pandemic came to the fore and there during the late summer there were fewer links than during the late winter.
4.1 Aggregate mobility
As we saw above, the data show that big box stores like Target and Walmart appear to have weathered the pandemic well, but the shift to remote work should also appear in the data. We can look at visits to the Comcast Center and the Plaza below it; visits in April and May, as the coronavirus took hold in the city, fell substantially.
4.2 Business in focus
Yet with offices vacant, parks should have swelled with visitors. We see mixed evidence of this in the data. Philadelphia has four central squares—Rittenhouse, Washington, Logan, and Franklin—which provide important community amenity; all saw fewer visits in April and May than later in the summer, suggesting winter patterns continued even as the weather improved. As a signal for tourism, we can look at Reading Terminal Market; vendors between its walls saw marked declines in visits beginning in April.
4.3 Tourism in focus
5 Detecting trends
In order to understand trends we fit a rolling average to these neighborhoods and plot these trends for context. The many neighborhoods in the Northeast and Northwest are rebound while the axis of University, Center and Old City are still down substantially.
5.1 Rolling averages
Appendix
A.1
| Best | Worst |
|---|---|
 |